In most companies the help centre belongs to nobody. It was written once, by whoever had time, and it is updated when a complaint gets loud enough. Meanwhile it is quietly the highest-traffic surface support owns, and every weak article in it generates tickets on a schedule.
I wrote and maintained that material for three markets for years. Treating it as documentation is what keeps it bad. Treating it as infrastructure is what makes it work.
Write from the user's words, not the internal taxonomy
Internally the feature has a name. Users do not know that name and will never search for it. The single most useful input I ever had was the raw text of what people actually typed — into search, into the contact form, into the first line of a ticket. That is the index. Titles should be the question as a person would ask it, not the feature as the org chart names it.
One article, one job
The article that tries to cover a topic comprehensively helps nobody, because the person reading it has one specific problem and is scanning. Splitting a long article into three narrow ones almost always reduces the tickets it was failing to prevent.
The measure is the ticket that never arrives
Which is an awkward metric, because it does not appear anywhere. The proxy I used: pick the five most common contact reasons, rewrite the material behind them, and watch those categories over the following weeks. It is slower than a dashboard and considerably more honest.
It rots without an owner
Products change and articles do not follow on their own. Without a named owner and a fixed review rhythm, the help centre degrades from asset to liability in about a year — and the failure is silent, because a wrong article does not error. It just sends someone confidently in the wrong direction, and then into the queue.